Skip to content

Security: qingsongcui/ContextCore

Security

SECURITY.md

Security Policy

ContextCore is currently designed for local-first, single-user desktop usage.

Supported Runtime Model

  • Backend and MCP endpoints bind to 127.0.0.1 by default.
  • REST and MCP endpoints do not yet implement login/session authentication.
  • The local threat model assumes you do not expose port 8420 to a LAN or the public internet.
  • Any local process on the same machine may be able to call the local API if it can reach 127.0.0.1:8420.

Agent Write-Back Model

ContextCore is safe by default:

  • Agents can read context and propose drafts.
  • Official knowledge writes, delete requests, and stale marks require user review by default.
  • Direct writes/deletes are advanced opt-in behavior and require both the Agent permission and knowledge-type policy to be set to allow.

Secrets

  • Do not commit .env files or real API keys.
  • CONTEXTCORE_OPENCLAW_BRIDGE_SECRET can be set to pin the OpenClaw bridge secret. If unset, the app generates a random secret for the process.
  • Optional OpenAI-compatible keys are used only for LLM/embedding features. Core local search and MCP access continue without them.

Reporting Issues

Please open a private report if the repository is private, or a GitHub Security Advisory if the repository is public. Include:

  • ContextCore version or commit hash.
  • Operating system.
  • Whether the issue affects REST API, MCP tools, backup/restore, local files, or desktop packaging.
  • Minimal reproduction steps.

There aren't any published security advisories